Integrative Analysis of ENCODE Data

When to Use

• User wants to combine multiple ENCODE experiments for cross-dataset analysis
• User asks about "integrating", "combining", or "comparing" experiments
• User wants to overlay histone marks with accessibility or expression data
• User needs to plan a multi-omic analysis using ENCODE data
• User asks about peak overlap, differential binding, or signal correlation
• User wants to perform ChromHMM segmentation using ENCODE histone data

Help the user combine multiple ENCODE experiments for cross-dataset or multi-omic analysis. This skill covers the full integration workflow: from defining the question and selecting compatible experiments, through choosing the right integration strategy and tools, to validating results and documenting provenance.

Literature Foundation

Step 1: Define the Integration Question

Clarify with the user which type of integration they need. There are four fundamental designs:

Each design has different requirements for compatibility, normalization, and statistical framework. Establish the design before searching for data.

Questions to ask the user:

• What biological question are you trying to answer?
• Are you comparing across samples (differential) or combining across samples (cataloging)?
• How many conditions/tissues/time points?
• Do you need statistical testing or descriptive overlap?

Step 2: Find Compatible Experiments

2a. Explore Data Availability

Start with encode_get_facets to understand what data exists before committing to a design:

encode_get_facets(
    assay_title="Histone ChIP-seq",
    organ="pancreas"
)

This returns counts by target, biosample, lab, and other facets. Use it to verify that the intended comparison has sufficient data on both sides.

2b. Search for Candidate Experiments

Search for experiments matching each arm of the integration:

encode_search_experiments(
    assay_title="Histone ChIP-seq",
    target="H3K27ac",
    organ="pancreas",
    biosample_type="tissue",
    limit=100
)

For multi-omic designs, search each assay layer separately:

# Accessibility layer
encode_search_experiments(assay_title="ATAC-seq", organ="pancreas", limit=50)

# Expression layer
encode_search_experiments(assay_title="total RNA-seq", organ="pancreas", limit=50)

# Histone layer
encode_search_experiments(assay_title="Histone ChIP-seq", target="H3K27ac", organ="pancreas", limit=50)

Present a summary table to the user showing experiments found per arm, number of replicates, labs represented, and any audit flags.

Step 3: Check Pairwise Compatibility

Track candidate experiments and then check compatibility:

encode_track_experiment(accession="ENCSR...")
encode_track_experiment(accession="ENCSR...")

encode_compare_experiments(
    accession1="ENCSR...",
    accession2="ENCSR..."
)

The compatibility check evaluates:

Critical rule: ALL experiments in an integration MUST share the same genome assembly. Never mix GRCh38 and hg19 coordinates without explicit liftOver.

Step 4: Select Matched Files

For each experiment, retrieve files using encode_list_files:

encode_list_files(
    experiment_accession="ENCSR...",
    file_format="bed",
    output_type="IDR thresholded peaks",
    assembly="GRCh38",
    preferred_default=True
)

File Matching Rules

All files entering the same integration MUST be matched on:

1. Same assembly (GRCh38 for human, mm10 for mouse)
2. Same output type (e.g., all "IDR thresholded peaks" or all "fold change over control")
3. Same file format (all narrowPeak, all bigWig, all TSV)
4. Same pipeline version when possible (check ENCODE pipeline annotations)

File Type Compatibility Matrix

Not all file types can be directly integrated. This matrix shows which combinations are valid:

Cannot directly combine:

• Raw FASTQ with proces